🏆 Arena
Model vs model — A/B comparisons, automated judging.
🏆 Arena Impact8.5winner avgvs7.8field avg=+0.7vs avg
Leaderboard
| # | Model | Provider | Avg | W | L | Best | Matches |
|---|---|---|---|---|---|---|---|
| #1 | deepseek-v4.1-flash | deepseek | 8.0 | 34 | 56 | 9.9 | 90 |
| #2 | glm-5.3-flash | zai | 8.1 | 12 | 38 | 9.7 | 50 |
| #3 | mimo-v2.6-flash | opencode-go | 8.2 | 9 | 33 | 10.0 | 42 |
| #4 | mimo-v2.5 | xiaomi | 7.4 | 9 | 26 | 8.9 | 35 |
| #5 | gpt-6-luna | commandcode | 7.9 | 5 | 41 | 9.9 | 46 |
| #6 | muse-spark-1.3-contributor | opencode-go | 7.8 | 5 | 44 | 9.9 | 49 |
| #7 | claude-haiku-5-5 | claude-cli | 7.6 | 4 | 31 | 8.6 | 35 |
| #8 | ling-3.1-flash | tokenharbor | 7.4 | 3 | 10 | 8.6 | 13 |
| #9 | space-bunny-free | opencode-go | 9.3 | 2 | 4 | 9.9 | 6 |
| #10 | mimo-v2.5-pro | xiaomi | 8.4 | 2 | 2 | 9.3 | 4 |
| #11 | cline-pass/minimax-m3 | cline-pass | 7.6 | 2 | 7 | 9.5 | 9 |
| #12 | poolside/laguna-s-2.1-free | openrouter | 6.8 | 2 | 33 | 8.4 | 35 |
| #13 | cmd-codex | command-code | 9.8 | 1 | 0 | 9.8 | 1 |
| #14 | longcat-2.5-preview-free | opencode-go | 8.2 | 1 | 16 | 9.9 | 17 |
| #15 | minimax-m3.1-flash | opencode-go | 7.6 | 1 | 16 | 9.2 | 17 |
| #16 | beam-free | opencode-go | 7.4 | 1 | 24 | 8.6 | 25 |
| #17 | claude-sonnet-5-5 | claude-cli | 8.3 | 0 | 5 | 9.7 | 5 |
| #18 | longcat-2.0 | opencode-go | 8.2 | 0 | 4 | 8.5 | 4 |
| #19 | minimax-m3 | opencode-go | 8.1 | 0 | 5 | 9.4 | 5 |
| #20 | minimax-m3.1-flash-free | opencode-go | 8.1 | 0 | 1 | 8.1 | 1 |
| #21 | step-5-preview-free | opencode-go | 8.1 | 0 | 3 | 8.2 | 3 |
| #22 | hy3 | opencode-go | 7.8 | 0 | 31 | 9.6 | 31 |
| #23 | omen-alpha | opencode-go | 7.8 | 0 | 7 | 8.6 | 7 |
| #24 | stealth/glyph-cluster:free | commandcode | 7.5 | 0 | 18 | 8.6 | 18 |
| #25 | gemini-3.8-flash-medium | agy | 7.2 | 0 | 14 | 8.5 | 14 |
| #26 | poolside-laguna-s-2.1-free | commandcode | 7.1 | 0 | 1 | 7.1 | 1 |
| #27 | qwen3.8-flash | tokenharbor | 7.1 | 0 | 8 | 8.4 | 8 |
| #28 | cline-pass-minimax-m3 | unknown | 6.9 | 0 | 1 | 6.9 | 1 |
| #29 | meituan/LongCat-2.0 | commandcode | 6.2 | 0 | 1 | 6.2 | 1 |
| #30 | ling-3.0-flash-sante | commandcode | 6.2 | 0 | 1 | 6.2 | 1 |
| #31 | qwen/qwen3.7-flash | openrouter | 4.7 | 0 | 2 | 7.7 | 2 |
Match History
▶AI-generated game assets in 2026: a four-stage sprite-to-soundtrack pipeline with real prices and Phaser audio wiringcontent2026-11-16deepseek-v4.1-flash9.2
#1deepseek-v4.1-flashcommandcode9.2
#2claude-haiku-5-5claude-cli8.5
#3gpt-6-lunacommandcode8.3
#4mimo-v2.6-flashopencode-go8.3
#5minimax-m3.1-flashopencode-go8.3
#6glm-5.3-flashopencode-go8.2
#7stealth/glyph-cluster:freecommandcode8.0
#8muse-spark-1.3-contributoropencode-go7.7
#9beam-freeopencode-go7.2
▶Weekly gaming news + trending games (9-14 Nov 2026)content2026-11-14gpt-6-luna7.8
#1gpt-6-lunacommandcode7.8
#2minimax-m3.1-flashopencode-go6.6
#3deepseek-v4.1-flashcommandcode6.5
#4muse-spark-1.3-contributoropencode-go6.4
#5beam-freeopencode-go6.0
#6mimo-v2.6-flashopencode-go5.6
#7claude-haiku-5-5claude-cli4.9
#8stealth/glyph-cluster:freecommandcode4.8
#glm-5.3-flashopencode-go0.0
#ling-3.1-flashtokenharbor0.0
▶Steam H1 2026 revenue record (Alinea estimate) and what indie browser-game devs can take from itnews-research2026-11-13claude-haiku-5-58.3
#1claude-haiku-5-5claude-cli8.3
#2glm-5.3-flashopencode-go8.2
#3gpt-6-lunacommandcode7.7
#4ling-3.1-flashtokenharbor7.4
#5deepseek-v4.1-flashcommandcode7.1
#6beam-freeopencode-go6.9
#7muse-spark-1.3-contributoropencode-go6.1
#claude-sonnet-5-5claude-code7.1
#mimo-v2.6-flashopencode-go0.0
#stealth/glyph-cluster:freecommandcode0.0
#minimax-m3.1-flashopencode-go0.0
▶Unity, Unreal 6, Godot, or Phaser: choosing a game engine for AI-assisted browser game development in 2026content2026-11-12ling-3.1-flash8.2
#1ling-3.1-flashtokenharbor8.2
#2mimo-v2.6-flashopencode-go8.1
#3gpt-6-lunacommandcode7.9
#4glm-5.3-flashopencode-go7.8
#5beam-freeopencode-go7.8
#6stealth/glyph-cluster:freecommandcode7.8
#7minimax-m3.1-flashopencode-go7.4
#8deepseek-v4.1-flashcommandcode7.2
#9claude-haiku-5-5claude-cli6.9
#muse-spark-1.3-contributoropencode-go0.0
▶nesformer: a working NES CPU compiled into Qwen3.5 weights, and what it means for game agentscontent2026-11-11glm-5.3-flash8.3
#1glm-5.3-flashopencode-go8.3
#2claude-haiku-5-5claude-cli8.2
#3minimax-m3.1-flashopencode-go7.7
#4stealth/glyph-cluster:freecommandcode7.6
#5deepseek-v4.1-flashcommandcode7.5
#6beam-freeopencode-go7.3
#7gpt-6-lunacommandcode7.3
#8mimo-v2.6-flashopencode-go7.3
#9muse-spark-1.3-contributoropencode-go0.0
#9ling-3.1-flashtokenharbor0.0
▶AI-Generated Beat Maps, the Honest Way: Build a Spec-Driven Rhythm Game in Phaser 3build2026-11-10glm-5.3-flash9.7
#1glm-5.3-flashopencode-go9.7
#2minimax-m3.1-flashopencode-go8.9
#3deepseek-v4.1-flashcommandcode8.9
#4mimo-v2.6-flashopencode-go8.8
#5gpt-6-lunacommandcode8.8
#6beam-freeopencode-go8.5
#7claude-haiku-5-5claude-cli8.5
#8muse-spark-1.3-contributoropencode-go8.3
#9stealth/glyph-cluster:freecommandcode0.0
#10ling-3.1-flashtokenharbor0.0
▶Build an Idle Tycoon Game in Phaser.js: A Beginner Tutorialbuild2026-11-09glm-5.3-flash9.0
#1glm-5.3-flashopencode-go9.0
#2gpt-6-lunacommandcode8.9
#3minimax-m3.1-flashopencode-go8.8
#4deepseek-v4.1-flashcommandcode8.6
#5beam-freeopencode-go8.6
#6claude-haiku-5-5claude-cli8.6
#7muse-spark-1.3-contributoropencode-go8.3
#8stealth/glyph-cluster:freecommandcode8.2
#9mimo-v2.6-flashopencode-go8.1
#10ling-3.1-flashtokenharbor0.0
▶Weekly Gaming News & Trends Nov 1–7 (Zelda OoT Switch 2, Pokémon UNITE sunset, Steam charts, releases)content2026-11-07muse-spark-1.3-contributor7.7
#1muse-spark-1.3-contributoropencode-go7.7
#2glm-5.3-flashopencode-go8.1
#3stealth/glyph-cluster:freecommandcode7.9
#4claude-haiku-5-5claude-cli7.8
#5deepseek-v4.1-flashcommandcode7.8
#6beam-freeopencode-go7.6
#7gpt-6-lunacommandcode7.5
#8mimo-v2.6-flashopencode-go7.3
#9minimax-m3.1-flashopencode-go6.5
#10ling-3.1-flashtokenharbor0.0
▶Phaser screen shake and hit feedback with camera shake and flashcontent2026-11-06glm-5.3-flash8.7
#1glm-5.3-flashopencode-go8.7
#2deepseek-v4.1-flashcommandcode8.6
#3muse-spark-1.3-contributoropencode-go8.6
#4mimo-v2.6-flashopencode-go8.4
#5gpt-6-lunacommandcode8.4
#6stealth/glyph-cluster:freecommandcode8.2
#7beam-freeopencode-go8.1
#8minimax-m3.1-flashopencode-go7.5
#9claude-haiku-5-5claude-cli7.1
#10ling-3.1-flashtokenharbor0.0
▶LLM-written room specs for procedural dungeon generation in Phasercontent2026-11-06claude-haiku-5-58.0
#1claude-haiku-5-5claude-cli8.0
#2deepseek-v4.1-flashcommandcode8.1
#3stealth/glyph-cluster:freecommandcode8.1
#4minimax-m3.1-flashopencode-go7.9
#5muse-spark-1.3-contributoropencode-go7.8
#6mimo-v2.6-flashopencode-go7.8
#7glm-5.3-flashopencode-go7.7
#gpt-6-lunacommandcode0.0
#beam-freeopencode-go0.0
#ling-3.1-flashtokenharbor0.0
▶Unreal Engine 6: What AI Game Developers Need to Know (queue aig-0019)content2026-11-05muse-spark-1.3-contributor7.1
#1muse-spark-1.3-contributoropencode-go7.1
#2stealth/glyph-cluster:freecommandcode8.0
#3deepseek-v4.1-flashcommandcode7.5
#4gpt-6-lunacommandcode6.8
#5beam-freeopencode-go6.8
#6minimax-m3.1-flashopencode-go6.5
#7claude-haiku-5-5claude-cli5.4
#8ling-3.1-flashtokenharbor5.3
#glm-5.3-flashopencode-go0.0
#mimo-v2.6-flashopencode-go0.0
▶Build a factory automation game in Phaser 3: grid-as-data tiles, a fixed tick loop, miner/belt/smelter/chest chain, throughput counter and upgrade loopbuild2026-11-04deepseek-v4.1-flash7.9
#1deepseek-v4.1-flashcommandcode7.9
#2gpt-6-lunacommandcode5.0
#3stealth/glyph-cluster:freecommandcode4.7
#4claude-haiku-5-5claude-cli4.5
#5muse-spark-1.3-contributoropencode-go4.4
#6glm-5.3-flashopencode-go4.1
#7beam-freeopencode-go4.1
#mimo-v2.6-flashopencode-go0.0
#minimax-m3.1-flashopencode-go0.0
#ling-3.1-flashtokenharbor0.0
▶Godot AI vs Godot-MCP vs Godot MCP Toolkit vs Godlink: which AI-agent bridge for Godot wins in 2026content2026-11-03ling-3.1-flash0.0
#0ling-3.1-flashtokenharbor0.0
#1deepseek-v4.1-flashcommandcode8.2
#2claude-haiku-5-5claude-cli8.1
#3minimax-m3.1-flashopencode-go7.6
#4glm-5.3-flashopencode-go7.4
#5muse-spark-1.3-contributoropencode-go7.2
#6stealth/glyph-cluster:freecommandcode7.2
#7gpt-6-lunacommandcode6.5
#8beam-freeopencode-go6.5
#9mimo-v2.6-flashopencode-go5.8
▶Build a Meccha Chameleon-style camouflage hide-and-seek game in Phaser 3content2026-11-02minimax-m3.1-flash9.2
#1minimax-m3.1-flashopencode-go9.2
#2glm-5.3-flashopencode-go8.8
#3mimo-v2.6-flashopencode-go8.7
#4muse-spark-1.3-contributoropencode-go8.6
#5claude-haiku-5-5claude-cli8.6
#6stealth/glyph-cluster:freecommandcode8.6
#7deepseek-v4.1-flashcommandcode8.5
#8beam-freeopencode-go8.5
#9gpt-6-lunacommandcode8.1
#10ling-3.1-flashtokenharbor8.0
▶Weekly gaming news + trending games (Oct 26-31, 2026)content2026-10-31mimo-v2.6-flash6.9
#1mimo-v2.6-flashopencode-go6.9
#2gpt-6-lunacommandcode6.9
#3hy3opencode-go6.8
#4muse-spark-1.3-contributoropencode-go6.6
#5glm-5.3-flashopencode-go6.6
#6deepseek-v4.1-flashcommandcode6.3
#7stealth/glyph-cluster:freecommandcode6.3
#8beam-freeopencode-go6.0
#9qwen3.8-flashtokenharbor5.9
#10minimax-m3.1-flashopencode-go5.1
#11gemini-3.8-flash-mediumagy4.5
#12ling-3.1-flashtokenharbor4.2
#13claude-haiku-5-5claude-cli4.2
#14poolside/laguna-s-2.1-freecommandcode0.0
▶Building a Donkey Kong-Style Barrel Climber in Phaser 3: Sloped Girders, Ladders and Rolling Barrelscontent2026-10-30gpt-6-luna8.4
#1gpt-6-lunacommandcode8.4
#2claude-haiku-5-5claude-cli8.4
#3claude-sonnet-5-5claude-code8.3
#4deepseek-v4.1-flashcommandcode8.0
#5stealth/glyph-cluster:freecommandcode7.7
▶How to Build a Chess Roguelike in Phaser.jscontent2026-10-30gpt-6-luna8.3
#1gpt-6-lunacommandcode8.3
#2gemini-3.8-flash-mediumagy7.9
#3glm-5.3-flashopencode-go7.9
#4deepseek-v4.1-flashcommandcode7.8
#5mimo-v2.6-flashopencode-go7.8
#6muse-spark-1.3-contributoropencode-go7.8
#7beam-freeopencode-go7.7
#8minimax-m3.1-flashopencode-go7.7
#9ling-3.1-flashtokenharbor7.5
#10claude-haiku-5-5claude-cli7.4
#11hy3opencode-go7.2
#12qwen3.8-flashtokenharbor7.1
#13stealth/glyph-cluster:freecommandcode6.7
▶Meshy vs Tripo vs Rodin: which AI text-to-3D tool should you use for game assets in 2026?content2026-10-29glm-5.3-flash8.4
#1glm-5.3-flashopencode-go8.4
#2stealth/glyph-cluster:freecommandcode8.4
#3deepseek-v4.1-flashcommandcode8.4
#4gpt-6-lunatokenharbor8.3
#5ling-3.1-flashtokenharbor8.3
#6mimo-v2.6-flashopencode-go8.2
#7hy3opencode-go8.0
#8claude-haiku-5-5commandcode7.9
#9beam-freeopencode-go7.9
#10muse-spark-1.3-contributoropencode-go7.8
#11minimax-m3.1-flashopencode-go7.6
#12qwen3.8-flashtokenharbor7.3
#13poolside/laguna-s-2.1-freecommandcode5.4
▶Phaser Camera Follow, Explained: Lerp, Deadzone and roundPixelscontent2026-10-28deepseek-v4.1-flash8.6
#1deepseek-v4.1-flashcommandcode8.6
#2ling-3.1-flashtokenharbor8.6
#3mimo-v2.6-flashopencode-go8.5
#4claude-haiku-5-5claude-cli8.4
#5gpt-6-lunacommandcode8.4
#6glm-5.3-flashopencode-go8.2
#7stealth/glyph-cluster:freecommandcode8.1
#8minimax-m3.1-flashopencode-go8.0
#9qwen3.8-flashtokenharbor7.8
#10beam-freeopencode-go7.8
#11gemini-3.8-flash-mediumagy7.7
#12muse-spark-1.3-contributoropencode-go7.6
#13hy3opencode-go7.5
#14poolside/laguna-s-2.1-freecommandcode6.2
▶Sloyd AI Review 2026: is the parametric AI 3D generator worth it for a beginner game developer?content2026-10-27ling-3.1-flash8.5
#1ling-3.1-flashtokenharbor8.5
#2mimo-v2.6-flashopencode-go8.2
#3claude-haiku-5-5claude-cli8.0
#4stealth/glyph-cluster:freecommandcode8.0
#5muse-spark-1.3-contributoropencode-go8.0
#6deepseek-v4.1-flashcommandcode8.0
#7gpt-6-lunacommandcode7.8
#8qwen3.8-flashtokenharbor7.6
#9glm-5.3-flashopencode-go7.6
#10beam-freeopencode-go7.6
#11minimax-m3.1-flashopencode-go7.5
#12hy3opencode-go7.3
#13gemini-3.8-flash-mediumagy6.5
▶Phaser 3 Game Over and Restart with Scene Flow: A Beginner Guidecontent2026-10-26ling-3.1-flash8.3
#1ling-3.1-flashtokenharbor8.3
#2beam-freeopencode-go8.3
#3gemini-3.8-flash-mediumagy8.2
#4poolside/laguna-s-2.1-freecommandcode8.2
#5deepseek-v4.1-flashcommandcode8.2
#6hy3opencode-go7.9
#7omen-alphaopencode-go7.8
#8glm-5.3-flashopencode-go7.8
#9muse-spark-1.3-contributoropencode-go7.4
#10mimo-v2.6-flashopencode-go7.3
#11claude-haiku-5-5claude-cli7.1
#12qwen3.8-flashtokenharbor6.8
#13gpt-6-lunacommandcode6.3
▶Weekly Gaming News and Trending Games: Week of 24 October 2026content2026-10-24muse-spark-1.3-contributor7.1
#1muse-spark-1.3-contributoropencode-go7.1
#2gpt-6-lunacommandcode6.8
#3claude-haiku-5-5claude-cli6.6
#4omen-alphaopencode-go6.5
#5beam-freeopencode-go6.4
#6ling-3.1-flashtokenharbor6.2
#7qwen3.8-flashtokenharbor5.9
#8mimo-v2.6-flashopencode-go5.8
#9glm-5.3-flashopencode-go5.5
#10gemini-3.8-flash-mediumagy5.1
▶Rebuilding Galaga in One File with Phaser 3: Formations, Dive-Bombs, and the Tractor Beambuild2026-10-23deepseek-v4.1-flash8.7
#1deepseek-v4.1-flashcommandcode8.7
#2claude-sonnet-5-5claude-code8.6
#3longcat-2.5-preview-freeopencode-go8.5
#4glm-5.3-flashopencode-go8.3
#5claude-haiku-5-5claude-cli8.3
#6gpt-6-lunacommandcode7.9
#7hy3opencode-go7.6
#8muse-spark-1.3-contributoropencode-go7.5
#9gemini-3.8-flash-mediumagy7.5
#10beam-freeopencode-go7.4
▶Retro Diffusion vs PixelLab: which AI pixel art generator a beginner should usecontent2026-10-22glm-5.3-flash8.0
#1glm-5.3-flashopencode-go8.0
#2deepseek-v4.1-flashcommandcode8.1
#3ling-3.1-flashtokenharbor8.0
#4beam-freeopencode-go7.7
#5gpt-6-lunacommandcode7.7
#6claude-haiku-5-5claude-cli7.5
#7mimo-v2.6-flashopencode-go7.4
#8gemini-3.8-flash-mediumagy7.2
#9hy3opencode-go6.8
#10muse-spark-1.3-contributoropencode-go5.9
▶Typewriter Text Dialogue Box in Phaser 3: Build a Story Box That Types Out Character by Charactercontent2026-10-21deepseek-v4.1-flash8.6
#1deepseek-v4.1-flashcommandcode8.6
#2claude-haiku-5-5claude-cli8.5
#3beam-freeopencode-go8.5
#4qwen3.8-flashtokenharbor8.4
#5gemini-3.8-flash-mediumagy8.3
#6hy3opencode-go8.2
#7mimo-v2.6-flashopencode-go8.2
#8minimax-m3.1-flash-freeopencode-go8.1
#9longcat-2.5-preview-freeopencode-go8.0
#10muse-spark-1.3-contributoropencode-go7.9
#11ling-3.1-flashtokenharbor7.7
#12gpt-6-lunacommandcode7.5
#13glm-5.3-flashopencode-go7.4
▶Coyote Time and Input Buffering: Make Your Phaser Jumps Feel Faircontent2026-10-19cline-pass/minimax-m38.8
#1cline-pass/minimax-m3cline-pass8.8
#2mimo-v2.6-flashopencode-go8.8
#3deepseek-v4.1-flashcommandcode8.4
#4beam-freeopencode-go8.4
#5glm-5.3-flashopencode-go8.2
#6claude-haiku-5-5claude-cli8.1
#7hy3opencode-go8.1
#8longcat-2.5-preview-freeopencode-go8.0
#9muse-spark-1.3-contributoropencode-go7.8
#10gemini-3.8-flash-mediumagy7.4
#11gpt-6-lunacommandcode7.2
▶Weekly gaming news and trending games: week of 17 October 2026content2026-10-17beam-free7.9
#1beam-freeopencode-go7.9
#2mimo-v2.6-flashopencode-go7.9
#3claude-haiku-5-5claude-cli7.6
#4muse-spark-1.3-contributoropencode-go7.4
#5cline-pass/minimax-m3cline-pass7.3
#6glm-5.3-flashopencode-go7.2
#7gpt-6-lunacommandcode6.8
#8longcat-2.5-preview-freeopencode-go6.2
#9gemini-3.8-flash-mediumagy6.2
#10meituan/LongCat-2.0commandcode6.2
#11hy3opencode-go5.8
#12deepseek-v4.1-flashcommandcode4.3
▶Phaser 3 Object Pooling: A Beginner's Bullet and Particle Poolcontent2026-10-16mimo-v2.6-flash8.7
#1mimo-v2.6-flashopencode-go8.7
#2gemini-3.8-flash-mediumagy8.5
#3claude-haiku-5-5claude-cli8.4
#4omen-alphaopencode-go8.3
#5muse-spark-1.3-contributoropencode-go8.0
#6hy3opencode-go8.0
#7deepseek-v4.1-flashcommandcode8.0
#8glm-5.3-flashopencode-go7.8
#9gpt-6-lunacommandcode7.8
#10longcat-2.5-preview-freeopencode-go7.7
#11cline-pass/minimax-m3cline-pass7.6
▶LDtk vs Tiled vs Ogmo Editor: Which Level Editor Should a Browser Game Beginner Use in 2026?content2026-10-15mimo-v2.6-flash8.7
#1mimo-v2.6-flashopencode-go8.7
#2muse-spark-1.3-contributoropencode-go8.6
#3omen-alphaopencode-go8.6
#4claude-haiku-5-5claude-cli8.4
#5poolside/laguna-s-2.1-freecommandcode8.3
#6gpt-6-lunacommandcode8.2
#7longcat-2.5-preview-freeopencode-go8.2
#8glm-5.3-flashopencode-go8.2
#9hy3opencode-go8.1
#10gemini-3.8-flash-mediumagy7.8
#11deepseek-v4.1-flashcommandcode7.0
#12cline-pass/minimax-m3cline-pass6.9
▶Gun Fight: Build a Two-Player Cover Shooter in Phaser 3 (Step 5 of 5)content2026-10-14mimo-v2.6-flash8.7
#1mimo-v2.6-flashopencode-go8.7
#2gemini-3.8-flash-mediumagy8.5
#3omen-alphaopencode-go8.3
#4claude-haiku-5-5claude-cli8.2
#5longcat-2.5-preview-freeopencode-go8.1
#6deepseek-v4.1-flashcommandcode8.0
#7glm-5.3-flashopencode-go7.9
#8minimax-m3cline-pass7.9
#9gpt-6-lunacommandcode7.5
#10muse-spark-1.3-contributoropencode-go7.5
#11poolside-laguna-s-2.1-freecommandcode7.1
#hy3opencode-go0.0
▶PixelLab: AI pixel-art asset generator reviewcontent2026-10-13mimo-v2.6-flash8.7
#1mimo-v2.6-flashopencode-go8.7
#2longcat-2.5-preview-freeopencode-go8.4
#3claude-haiku-5-5claude-cli8.4
#4hy3opencode-go8.3
#5muse-spark-1.3-contributoropencode-go8.2
#6deepseek-v4.1-flashcommandcode8.0
#7omen-alphaopencode-go7.8
#8glm-5.3-flashopencode-go7.5
#9gpt-6-lunacommandcode7.0
#10poolside/laguna-s-2.1-freecommandcode7.0
#11cline-pass-minimax-m3unknown6.9
▶Phaser scene flow for beginners: title, game, pause and game-over with this.scene and the Registryguide2026-10-12claude-haiku-5-57.5
#1claude-haiku-5-5claude-cli7.5
#2deepseek-v4.1-flashcommandcode7.5
#3longcat-2.5-preview-freeopencode-go7.3
#4hy3opencode-go7.2
#5omen-alphaopencode-go7.2
#6gpt-6-lunacommandcode7.0
#7glm-5.3-flashopencode-go6.7
#8muse-spark-1.3-contributoropencode-go6.4
#9poolside/laguna-s-2.1-freecommandcode6.2
#10cline-pass/minimax-m3cline-pass5.8
#mimo-v2.6-flashopencode-go0.0
#step-5-preview-freeopencode-go0.0
▶Weekly gaming news + trending games (week of 10 Oct 2026): five news items with inline citations, Phaser/AI prototype ideas, FAQ, takeawaysnews2026-10-10cline-pass/minimax-m39.5
#1cline-pass/minimax-m3cline-pass9.5
#2deepseek-v4.1-flashcommandcode9.2
#3glm-5.3-flashopencode-go8.7
#4gpt-6-lunacommandcode8.6
#5mimo-v2.6-flashopencode-go8.5
#6muse-spark-1.3-contributoropencode-go8.4
#claude-haiku-5-5claude-cli0.0
#poolside/laguna-s-2.1-freecommandcode0.0
#hy3opencode-go0.0
#longcat-2.5-preview-freeopencode-go0.0
#meituan/LongCat-2.0commandcode0.0
#nvidia/nemotron-3-ultra-550b-a55bcommandcode0.0
#omen-alphaopencode-go0.0
#step-5-preview-freeopencode-go0.0
▶3D Gaussian Splatting for game developers in 2026 - a desk-research deep-dive (Friday research slot): what 3DGS is (Kerbl et al., SIGGRAPH 2023, arXiv:2308.04079; a splat is a translucent 3D blob, NOT a mesh), the 2026 format ecosystem (PLY source; PlayCanvas SOG announced 2025-09-17, 15-20x smaller than PLY, Skate Park 1 GB PLY -> 42 MB .sog ~95% reduction, Morton order, WebGPU compression, Engine 2.11.0, streamed SOG; Niantic SPZ MIT 'JPG for splats' ~10x, SPZ 4 2026-05-05 six parallel ZSTD streams, 8.5 GB/34M-pt encode 3m26s -> 1m8s ~3x, 2.5% smaller, 1.5-2.1x time-to-render, 10M-pt cap removed, ~800k Photoshop SPZ files; GLB + Khronos KHR_gaussian_splatting), engine/browser support (Three.js r186 NATIVE gsplat via WebGPU/TSL + SPZ loaders; PlayCanvas first-class gsplat; Unreal NO first-class Epic splat type - Luma AI Fab plugin hijacks the renderer via Niagara, ~2M particle cap so scenes must be chunked, no shadows, no ray tracing on translucent splats, wrong-order compositing; Unity aras-p plugin experimental; viewers antimatter15/splat, mkkellogg/GaussianSplats3D, SuperSplat, WebSplatter WebGPU-native ACM MM 2026 arXiv:2602.03207 wait-free hierarchical radix sort + opacity-aware culling, 1.2x-4.5x speedups), mesh extraction (SuGaR CVPR 2024 Poisson reconstruction ~30-35 min single GPU vs hours for neural-SDF, hybrid mesh+splat; 2D-SuGaR Eurographics 2026 arXiv:2605.00569), the capture->SOG/SPZ->load->optional-mesh pipeline, and the honest limitations (not meshes, no collision/skinning/LOD-as-mesh/standard materials, transparency sorting, GB-scale PLY, mobile memory ceiling, immature relighting, poor fit for authored animated characters, plugin-grade big-engine support). Desk research only - no tool run, scene built or scan performed (stated verbatim in the published post). Pipeline: plan = cline-pass/glm-5.3 (brief_contract true); research = orchestrator in-session web passes over primary sources (arXiv, PlayCanvas docs/blog, Niantic, Khronos, Fab, Three.js discourse); 13-lane canonical roster via parallel-builders.py - 10/13 returned, glm-5.3-flash / ling-3.0-flash-sante / cline-pass/qwen3.8-max DNF; all 10 drafts judged with qwen3.7-flash multipass (3 passes, mean-gated); winner deepseek-v4.1-flash 8.42 (least-noisy winner, 8.45/8.65/8.15); step-5-preview-free 8.13 and claude-haiku-5-5 8.07 also cleared 8.0. QA = claude-sonnet-5-5 via claude-cli (fixed the false 'Per the PlayCanvas docs' closing attribution, attributed SuGaR's Blender/Unity/Unreal tools, linked the three named GitHub viewers/plugins, dropped an unsourced 'artist weeks' claim); all 13+3 external links re-verified (12+3 = HTTP 200; the Fab listing returns 403 to datacenter IPs - live in browsers, carve-out). Hero = Lumenfall FLUX.1-schnell (r2 webp, HTTP 200, 605 KB).content2026-10-09deepseek-v4.1-flash8.4
#1deepseek-v4.1-flashcommandcode8.4
#2step-5-preview-freeopencode-go8.1
#3claude-haiku-5-5claude-cli8.1
#4muse-spark-1.3-contributoropencode-go8.0
#5mimo-v2.6-flashopencode-go8.0
#6longcat-2.5-preview-freeopencode-go8.0
#7poolside/laguna-s-2.1-freecommandcode7.4
#8hy3opencode-go7.3
#9cline-pass/minimax-m3cline-pass7.1
#10gpt-6-lunacommandcode7.0
▶Build a Lunar Lander game in Phaser 3 (Friday game build + blog post, 2026-10-09). Ships a playable single-file Phaser 3.60 prototype at /games/lunar-lander-phaser.html (960x540, Phaser.Scale.FIT, procedural terrain + lander textures via make.graphics().generateTexture, manual gravity/thrust/rotation integration, 1x/2x/5x pads, fuel bar, HUD, localStorage high score, R to restart, ?mode=bot autopilot) plus a beginner tutorial post. Angle: the 1979 Atari vector classic (Atari's first vector arcade game, Aug 1979, designers Wendi Allen + Rich Moore, 4,830 units, pads pay 1x/2x/5x) rebuilt for beginners - the post teaches the loop while labelling all rebuild physics constants as design choices, not ROM values. Pipeline: plan = cline-pass/glm-5.3 (brief_contract true); research = mimo-v2.6-pro via commandcode (992w dossier) plus orchestrator in-session web verification of every cited fact; 13-lane canonical roster via parallel-builders.py - 9/13 returned (glm-5.3-flash, ling-3.0-flash-sante, cline-pass/qwen3.8-max, longcat-2.5-preview-free DNF); all 9 drafts judged qwen3.7-flash multipass (3 passes, mean-gated) - 5 cleared 8.0. Winner claude-haiku-5-5 8.35. QA = claude-sonnet-5-5 via claude-cli + mechanical pass (all 17 external URLs HTTP 200, zero numbered refs, zero banned slop, no HERO_IMAGE_PROMPT leak, iframe embed present); fixed CENTER_BOTH->NO_CENTER to match the shipped file and a stale 'section 9' comment reference. Game render-gate (Playwright): zero pageerrors, canvas 960x540, thrust changes fuel/altitude, ?mode=bot lands (state landed), PNG non-blank + distinct md5s. Hero = Lumenfall FLUX.1-schnell (r2 webp, HTTP 200, 360 KB).game-post2026-10-09claude-haiku-5-58.3
#1claude-haiku-5-5claude-cli8.3
#2step-5-preview-freeopencode-go8.2
#3mimo-v2.6-flashopencode-go8.1
#4gpt-6-lunacommandcode8.1
#5deepseek-v4.1-flashcommandcode8.1
#6claude-sonnet-5-5claude-code8.0
#7cline-pass/minimax-m3cline-pass7.3
#8poolside/laguna-s-2.1-freecommandcode6.9
#9hy3opencode-go6.7
#10muse-spark-1.3-contributoropencode-go4.8
#glm-5.3-flashzai0.0
#ling-3.0-flash-santecommandcode0.0
#cline-pass/qwen3.8-maxcline-pass0.0
#longcat-2.5-preview-freeopencode-go0.0
▶Build a Dig Dug game in Phaser 3 (Friday game post, 2026-10-09). Ships a playable single-file Phaser 3.60 prototype at /games/dig-dug-style-phaser.html (480x488 canvas, 15x14 tile grid, Phaser.Scale.FIT, procedural textures via make.graphics().generateTexture, tile digging at 10 pts/tile, 4-step air-pump inflation with depth-band pop scoring 200/300/500, Pooka + Fygar enemies with horizontal fire, three falling rocks with 1000/2500/4000 crush scoring, bonus fruit after two rocks, 3 lives, HUD, localStorage high score, R to restart, ?mode=bot autopilot, touch D-pad) plus a beginner tutorial post. Angle: the 1982 Namco maze classic (designer Masahisa Ikegami with Galaga creator Shigeru Yokoyama, Namco Galaga board, Feb 20 1982 JP / April 1982 NA, second highest-grossing arcade game of 1982 in Japan) rebuilt for beginners; all rebuild tuning constants are labelled as design choices, not ROM values. Pipeline: plan = cline-pass/glm-5.3 (brief_contract true); research = mimo-v2.6-pro via commandcode (dossier rejected as hallucinated on credits/dates, so the orchestrator re-verified every fact against the Wikipedia article in-session); 13-lane canonical roster via parallel-builders.py - 10/13 returned (longcat-2.5-preview-free, ling-3.0-flash-sante, cline-pass/qwen3.8-max DNF after 3 attempts); all 10 drafts JEV-triaged band=solid and judged qwen3.7-flash multipass (3 passes, mean-gated) - 3 cleared 8.0. Winner gpt-6-luna 8.23. QA = claude-sonnet-5-5 via claude-cli (verdict FAIL; applied frontmatter + hero + arcade-museum citation, description trimmed to 156 chars; all 15 external URLs pre-verified HTTP 200 by the orchestrator - QA's own link check was approval-blocked). Game render-gate (Playwright): zero pageerror/console errors, canvas 480x488 scaled, keyboard input digs a tile and moves the player, ?mode=bot advances a level, PNG non-blank. Hero = Lumenfall FLUX.1-schnell (r2 webp, HTTP 200, 344 KB).game-post2026-10-09gpt-6-luna8.2
#1gpt-6-lunacommandcode8.2
#2mimo-v2.6-flashopencode-go8.1
#3glm-5.3-flashzai8.1
#4cline-pass/minimax-m3cline-pass7.9
#5step-5-preview-freeopencode-go7.9
#6deepseek-v4.1-flashcommandcode7.6
#7claude-haiku-5-5claude-cli7.5
#8muse-spark-1.3-contributoropencode-go7.3
#9hy3opencode-go7.1
#10poolside/laguna-s-2.1-freecommandcode4.3
#longcat-2.5-preview-freeopencode-go0.0
#ling-3.0-flash-santecommandcode0.0
#cline-pass/qwen3.8-maxcline-pass0.0
▶Phaser Game Agent review 2026 - a desk-research review of Phaser Studio's tool that turns a one-sentence prompt into a complete, playable browser game (code, art, sound). Covers the 2026-07-01 launch and the 2026-06-12 'active beta' blog; the Phaser AE ('Arcade Engine') verb-based API, capability-named blocks and the 'the engine never learns games. The library does' design line; the Claude-managed-agents + Superserve sandbox build loop; the two entry points (hosted agent at phaser.io/agent/about and the MCP server for Claude Code/Cursor/VS Code/Codex/Gemini CLI/Antigravity/Qoder/Windsurf); setup via npx @phaserjs/game-agent (Node 20+), free setup and build time from $0.01/min; credits - a fully featured game is ~500 credits, packs 3,500+250=$10 / 6,000+500=$15 / 10,500+2,500=$25 / 22,000+4,000=$50 / 44,000+8,000=$100 / 100,000+30,000=$250 (effective $0.00267-$0.00192 per credit); the 2026-09-22 model picker (GROK/Grok 4.7 4,000; MUSE/Muse Spark 1.3 5,000; SONNET/Claude Sonnet 5 5,000 default; OPUS/Claude Opus 5.5 10,000 - minimum balances, not cost) on Phaser AE v2.8.1, with AE v3 + a Game Design Agent announced but unreleased; export to Web/Windows/Linux/macOS plus TypeScript source-code export (local toolchain is the user's job); and the honest limitations (cloud/AE lock-in, upfront credit minimums, frozen per-game versions, WebGPU-only card effects). Desk research only - no game built, no account created, no credits spent, no MCP client installed (stated verbatim in the published post). Pipeline: plan by deepseek-v4-pro/commandcode; research = orchestrator in-session web passes over Phaser primary pages fetched 2026-10-06; canonical 10-lane builder array via parallel-builders.py - 8/10 returned, space-bunny-free and longcat-2.5-preview-free DNF; judge qwen3.7-flash multipass (3 passes, mean-gated); winner mimo-v2.6-flash 8.32 (only lane alongside glm-5.3-flash 8.15 to clear 8.0); the plan's literal 'Band:' lead marker was removed at QA. Hero = Lumenfall FLUX.1-schnell (r2 webp).content2026-10-06mimo-v2.6-flash8.3
#1mimo-v2.6-flashopencode-go8.3
#2glm-5.3-flashopencode-go8.2
#3deepseek-v4.1-flashcommandcode7.9
#4gpt-6-lunacommandcode7.8
#5poolside/laguna-s-2.1-freecommandcode7.8
#6muse-spark-1.3-contributoropencode-go7.8
#7hy3opencode-go7.6
#8ling-3.0-flash-santecommandcode6.2
▶Beginner guide to publishing a Phaser game in 2026: why a Phaser game cannot run from file:// (browser lock-down), how the official Phaser Vite template builds a dist/ folder (npm install / npm run dev / npm run build, upload ALL of dist/), and how the four free hosts compare for a beginner — itch.io (ZIP, <=1000 files, <=500MB total, <=200MB per file, HTTPS only), GitHub Pages (free for public repos, <=1GB site, soft 100GB/month, soft 10 builds/hour), Netlify Free (300 credits/month hard cap; 15 credits per production deploy, 20 credits/GB bandwidth), and Cloudflare Pages Free (500 builds/month, 20,000 files, 25 MiB per file, no bandwidth cap listed), plus Vercel Hobby / GitLab Pages as FAQ honorable mentions — with a decision table and a which-should-you-pick rule. Desk research only: every figure read from the vendors' current official docs/pricing pages on 2026-10-05; no game deployed, no account created, no upload performed. 10-lane canonical builder array; ling-3.0-flash-sante DNF (empty commandcode stream); 9 drafts judged with qwen3.7-flash multipass (3 passes, mean-gated) behind a measurement-carrying verified-fact preamble; winner muse-spark-1.3-contributor 9.9, clear of the 9.87 runner-up.content2026-10-05muse-spark-1.3-contributor9.9
#1muse-spark-1.3-contributoropencode-go9.9
#2gpt-6-lunacommandcode9.9
#3longcat-2.5-preview-freeopencode-go9.9
#4space-bunny-freeopencode-go9.8
#5hy3opencode-go9.6
#6deepseek-v4.1-flashcommandcode9.5
#7mimo-v2.6-flashopencode-go9.0
#8glm-5.3-flashopencode-go8.3
#9poolside/laguna-s-2.1-freecommandcode6.6
▶Steam-like redesign of the /games/ hub: featured spotlight + horizontal genre/engine shelves + dense Full Librarydesign-spec2026-10-04deepseek-v4.1-flash9.8
#1deepseek-v4.1-flashcommandcode9.8
#2claude-sonnet-5-5claude-cli9.7
#3muse-spark-1.3-contributoropencode-go9.7
#4glm-5.3-flashopencode-go9.5
#5mimo-v2.6-flashopencode-go9.5
#6hy3opencode-go9.4
#7longcat-2.5-preview-freeopencode-go9.3
#8ling-3.0-flash-santecommandcode0.0
#9gpt-6-lunacommandcode0.0
#10space-bunny-freeopencode-go0.0
#11poolside/laguna-s-2.1-freecommandcode0.0
▶Tile-based level design in Phaser 3 (3.90) with the Tiled editor - Monday beginner guide. Canonical Phaser Tiled recipe (load.tilemapTiledJSON -> make.tilemap -> addTilesetImage -> createLayer -> setCollisionByProperty -> physics.add.collider -> camera/world bounds -> createFromObjects) plus the three AI-assistant mistakes to watch for: removed pre-3.50 createStaticLayer/createDynamicLayer, addTilesetImage argument order (Tiled tileset NAME vs image KEY), and Phaser's Tiled parser not supporting 'Collection of Images' tilesets. Desk research only from official Phaser + Tiled docs (Phaser 3.90, Tiled 1.12.2) - no hands-on testing. 10-lane canonical builder array via parallel-builders.py; space-bunny-free and longcat-2.5-preview-free DNF (no output); ling-3.0-flash-sante returned a truncated 179-word draft with no HERO_IMAGE_PROMPT (ineligible); 7 drafts judged with qwen/qwen3.7-flash multipass (3 passes, mean-gated); winner gpt-6-luna 9.0 (five drafts >= 8.0).content2026-10-04gpt-6-luna9.0
#1gpt-6-lunacommandcode9.0
#2glm-5.3-flashopencode-go8.8
#3mimo-v2.6-flashopencode-go8.7
#4muse-spark-1.3-contributoropencode-go8.6
#5deepseek-v4.1-flashcommandcode8.5
#6hy3opencode-go7.8
#7poolside/laguna-s-2.1-freecommandcode7.6
#8ling-3.0-flash-santecommandcode0.0
#9space-bunny-freeopencode-go0.0
#10longcat-2.5-preview-freeopencode-go0.0
▶AI playtesting agents for browser games in 2026: how LLM and vision-language agents test games. TITAN LLM MMORPG testing (95% task completion, 4 previously-unknown bugs, 8 real QA pipelines), the human-AI VLM study (800 test cases, 276 participants; GPT-4o 87.5% scene-description / 97.1% defect; 62.7% AI-assisted vs 41.3% manual, p<0.001; 64.4% knowledge+AI), Razer QA Companion-AI (GDC 2026) and modl.ai, the ysskrishna 2048 vision+Playwright+LangGraph harness, and the pixel-vs-ARIA two-surface testing idea. Desk research only - no hands-on testing. Friday research deep-dive slot.content2026-10-04space-bunny-free9.9
#1space-bunny-freeopencode-go9.9
#2muse-spark-1.3-contributoropencode-go9.9
#3mimo-v2.6-flashopencode-go9.8
#deepseek-v4.1-flashcommandcode0.0
#glm-5.3-flashopencode-go0.0
#hy3opencode-go0.0
#gpt-6-lunacommandcode0.0
#longcat-2.5-preview-freeopencode-go0.0
#poolside/laguna-s-2.1-freecommandcode0.0
▶A* pathfinding for enemy AI in a Phaser 3 grid game — beginner tutorial (Wednesday slot). Angle: represent a tile level as a graph, contrast BFS / Dijkstra / Greedy Best-First / A*, implement A* in plain JavaScript (g, h, f = g + h, came_from, sorted-array open set), wire it into a Phaser 3 Scene so the player moves with arrow keys and an enemy re-paths and chases via lerp, then tune heuristics and smoothing and shrink the graph. Desk research only — no build shipped or playtested (stated verbatim in the post). 10-lane builder roster dispatched via parallel-builders.py; 7 returned, 3 DNF (space-bunny-free, longcat-2.5-preview-free, ling-3.0-flash-sante — pool capped at 420s); 4 drafts judged with qwen3.7-flash multipass (3 passes, mean-gated); winner glm-5.3-flash 8.45 (gpt-6-luna 8.27 also passed). Research = orchestrator in-session web passes; plan = deepseek-v4-pro via commandcode.content2026-10-04glm-5.3-flash8.4
#1glm-5.3-flashopencode-go8.4
#2gpt-6-lunacommandcode8.3
#3hy3opencode-go7.7
#4muse-spark-1.3-contributoropencode-go7.3
▶Convai vs Inworld vs DIY: how to add talking NPCs to a browser game in 2026 (comparison slot). Angle: the 2026 decision is a 'who owns the character' decision. Convai still runs a full hosted conversational-NPC platform (STT + LLM + TTS + text-to-action + lip sync, per-character knowledge bank, long-term memory, Vision, Actions; tiers Free $0 / Indie Dev $29/mo / Professional $99/mo / Scale $499/mo / Business $1,199/mo, Enterprise custom; an 'interaction' is one input + one response and quotas do not roll over; WebGL is full for core NPC features but no spatial audio) whereas Inworld RETIRED its hosted Character Studio between March and September 2025 and now sells realtime TTS, STT, a Realtime API (STT+LLM+TTS in one WebSocket, OpenAI Realtime protocol compatible) and a 220+ LLM Router (On-Demand free, Creator $25 / Builder $100 / Developer $300 / Growth $1,500 per month, LLMs at provider cost) so the developer owns the character. The third path is a DIY browser loop (MDN Web Speech API: SpeechSynthesis Baseline since Sep 2018, SpeechRecognition limited/not Baseline and server-backed on Chrome) where you pay only LLM token cost. Desk research only: every price, limit and feature was read from the vendors' own pricing/docs pages on 2026-10-04 (orchestrator in-session web passes; all 17 candidate URLs HTTP 200) and no tool was run hands-on. 10-lane canonical roster dispatched via parallel-builders.py; 8 returned, 2 DNF (space-bunny-free and ling-3.0-flash-sante; pool capped at 420s); all 8 drafts judged with qwen3.7-flash multipass (3 passes, mean-gated) behind a measurement-carrying verified-fact preamble; winner longcat-2.5-preview-free 9.50; 7 of 8 drafts >= 8.0. Plan = deepseek-v4-pro via commandcode (brief_contract true). QA = claude-sonnet-5-5 via qa-glm53.py, which returned a TRUNCATED article (305 words) so QA was completed deterministically (all 12 unique links re-verified HTTP 200; 7/7 H2 leads measured 41-48 words; zero em dashes; zero numbered refs; stray meta line moved to frontmatter). Hero = Lumenfall FLUX.1-schnell (r2 webp, HTTP 200, 343 KB) after one failed+Lumenfall/HF pass and one successful retry.content2026-10-04longcat-2.5-preview-free9.5
#1longcat-2.5-preview-freeopencode-go9.5
#2mimo-v2.6-flashopencode-go9.3
#3muse-spark-1.3-contributoropencode-go9.0
#4deepseek-v4.1-flashcommandcode9.0
#5gpt-6-lunacommandcode8.8
#6glm-5.3-flashopencode-go8.3
#7hy3opencode-go8.1
#8poolside/laguna-s-2.1-freecommandcode5.4
▶Tripo AI Review 2026: is the fast text-to-3D / image-to-3D generator ready for a game developer's asset pipeline? (Tuesday tool-review slot; desk research only)content2026-10-04deepseek-v4.1-flash8.7
#1deepseek-v4.1-flashcommandcode8.7
#2mimo-v2.6-flashopencode-go8.2
#3longcat-2.0opencode-go8.2
#4longcat-2.5-preview-freeopencode-go8.2
#5minimax-m3opencode-go8.1
#6glm-5.3-flashopencode-go7.9
#7gpt-6-lunacommandcode7.8
#8muse-spark-1.3-contributoropencode-go7.6
#9hy3opencode-go7.3
#10space-bunny-freeopencode-go0.0
#11pixel-canarycommandcode0.0
#12ling-3.0-flash-fin-freecommandcode0.0
▶Weekly AI game-dev news roundup, week of September 27 - October 3, 2026 (Saturday news slot). Five new stories, deduped against the Sep 5 / Sep 19 / Sep 26 roundups: (1) AMD to acquire World Labs for $8.2B - Fei-Fei Li's world-model startup whose Marble tool exports Gaussian-splat 3D spaces for game development; (2) GamesIndustry.biz HR Summit (London, Oct 1) - AI is creating more work, not less; (3) Bose acquires Firelight Technologies, maker of FMOD interactive audio; (4) Grasshopper Manufacture splits from NetEase and goes independent; (5) EU Consumer Protection Cooperation Network opens coordinated actions against nine games companies over in-game currency. Desk research only on primary sources - no hands-on testing. Pipeline: research = orchestrator in-session web passes (mimo-v2.6-pro/commandcode research lane is unreliable on this blog); plan = deepseek-v4-pro via commandcode; 8-lane builder array via parallel-builders.py, 8/8 drafts returned; judge = qwen/qwen3.7-flash 3-pass multipass on a verified-fact preamble, stopped at first >=8.0 (mimo 9.47). Hero = Lumenfall FLUX.1-schnell (r2 webp, HEAD 200).content2026-10-03mimo-v2.6-flash9.5
#1mimo-v2.6-flashopencode-go9.5
#2deepseek-v4.1-flashcommandcode9.3
#3longcat-2.5-preview-freeopencode-go7.0
#4glm-5.3-flashopencode-go0.0
#5muse-spark-1.3-contributoropencode-go0.0
#6hy3opencode-go0.0
#7gpt-6-lunacommandcode0.0
#8space-bunny-freeopencode-go0.0
▶Build a Bomberman-style grid bomber in Phaser 3 (Friday 2026-10-02 game build + beginner tutorial post). Ships a playable single-file Phaser 3.90.0 prototype at /games/bomberman-style-grid-bomber-phaser.html plus a beginner tutorial. Fixed spec: 15x13 grid, 32px tiles, 480x416 playfield + 48px HUD = 480x464 canvas, Scale.FIT + CENTER_BOTH, procedural textures only, no physics engine (logical 2-D grid), one tile per keypress with a 140ms tween and an isMoving lock, arrow keys + WASD via JustDown, Space places bombs capped by bombCount (starts 1), 3.0s fuse, cross blast range 2, chain detonation, flames ~450ms, power-ups +1 bomb / +1 range, exit under one random brick, 3 lives, 3 levels, swipe + on-screen d-pad/bomb button. The prototype was smoke-tested in a headless Chromium probe: 10/10 assertions (no console/page errors, canvas 480x464 scaled, keyboard movement, hard-wall blocking, bomb placement then fuse then clear, a soft brick destroyed, distinct rendered frames).game-post2026-10-02mimo-v2.6-flash10.0
#1mimo-v2.6-flashopencode-go10.0
#2deepseek-v4.1-flashcommandcode9.9
#3gpt-6-lunacommandcode9.8
#4space-bunny-freeopencode-go9.5
#5muse-spark-1.3-contributoropencode-go9.0
#6longcat-2.5-preview-freeopencode-go8.9
#7glm-5.3-flashopencode-go8.8
#8hy3opencode-go8.8
▶Build a Q*bert-style isometric hopper in Phaser 3 (Friday 2026-10-02 game build + beginner tutorial, the second of the day's two game jobs). Ships a playable single-file Phaser 3.90.0 prototype at /games/qbert-phaser.html plus a beginner tutorial. Fixed spec: 28-cube isometric pyramid (cells where col+row<=6), 560x640 canvas, Scale.FIT + CENTER_BOTH, 2:1 dimetric diamond tiles drawn with Graphics.fillPoints, snap-to-grid hop tweens (150ms) with an isMoving lock, arrow keys + WASD and four screen-quadrant touch zones, per-cube hitCount states (one-hit then two-hit levels) cleared across all 28 cubes, Coily chaser + Slick/Sam cube-reverters + red balls, two edge discs that teleport to the apex, 3 lives, localStorage hi-score, game-over on a void hop. Smoke-tested in a headless-Chromium probe: 27/27 assertions pass (0 console/page errors beyond a favicon 404, canvas 560x640 scaled, diagonal hop + score, mid-air input lock, fall = -1 life, disc -> apex, forced level clear, game over, 3 distinct non-blank frames, mobile 390x844 the canvas fits and a touch zone triggers a hop); re-verified against the deployed URL after deploy.game-post2026-10-02mimo-v2.6-flash8.9
#1mimo-v2.6-flashopencode-go8.9
#2muse-spark-1.3-contributoropencode-go8.8
#3gpt-6-lunacommandcode8.1
#4longcat-2.5-preview-freeopencode-go8.0
#5hy3opencode-go8.0
#6glm-5.3-flashopencode-go7.8
#7deepseek-v4.1-flashcommandcode7.6
#space-bunny-freeopencode-go0.0
▶Make your Phaser 3 browser game playable on mobile (Wednesday tutorial slot). Angle: a desktop-oriented Phaser game becomes mobile-playable with five well-defined pieces — responsive scaling (Phaser.Scale.FIT + CENTER_BOTH, 800x600 world), unified pointer events (pointerdown/pointermove/pointerup, this.input.addPointer(2), disableContextMenu), a radius-clamped virtual joystick (base on pointerdown, thumb clamped to radius, Math.atan2 angle + clamped magnitude -> sprite.setVelocity), circular action-button hit areas (new Phaser.Geom.Circle(r,r,r) + Phaser.Geom.Circle.Contains), and two HTML/CSS fixes (viewport meta tag + canvas { touch-action: none }). Also covers scale resize / orientationchange events, isPortrait/isLandscape, setGameSize, and iOS audio unlock on first pointer event. HONEST DESK RESEARCH: grounded against official Phaser Scale Manager and MDN docs on 2026-09-30; no physical-device testing performed (post carries that disclosure). Every inline link verified HTTP 200 — two originally-404ing docs.phaser.io/api-documentation URLs were replaced in QA with verified-200 alternatives (concepts/scale-manager, api-documentation/class/geom-circle).content2026-09-30glm-5.3-flash8.6
#1glm-5.3-flashopencode-go8.6
#2gpt-6-lunacommandcode8.0
#3minimax-m3opencode-go6.9
▶Weekly AI game-dev news roundup, week of September 20-26, 2026 (Saturday news slot). Angle: the week AI game development got official on-ramps. Anchor stories: (1) Unity shipped first-party AI coding-agent plugins — Claude Code plugin live 2026-09-09, Codex plugin 2026-09-16 — bundling 29 Unity engineering skills incl. a unity-cli skill, the Unity CLI and Unity's MCP server for live Editor control, so external agents stop guessing at stale engine APIs (https://unity.com/blog/unity-plugin-for-claude-code); (2) Meta Connect 2026 announced Horizon Create (mobile) and Horizon Studio (browser), prompt-to-playable 2D/3D tools built on the Meta Horizon Engine, plus Meta VR Glasses (5x lighter than Quest 3, 5K micro-OLED, Spring 2027, $1,299.99) with Unity 6.6/OpenXR support from day one; (3) PUBG Ally (arXiv:2609.29837, submitted 2026-09-24), a voice-enabled embodied AI teammate trained on ~39k live sessions, ~1.6s on-device per spoken exchange vs 3.4s cloud, +25.1pp recommendation; (4) Owlchemy Labs + Artificial Agency shipped agentic NPCs (T.E.M.P.s) in Dimensional Double Shift using a small game-specific model; (5) arXiv:2609.02931 — a locally hosted Mistral 7B via Ollama steering a PPO policy, win rate 11%->24% vs a Balanced opponent but 'Surround' chosen 83.8% of the time; (6) Microsoft's 'Generative Narrative Game Experience With Player Feedback' patent (reported 2026-09-23); quick-links table covers Phaser 4.2/4.2.1 and Trophy Games acquiring Playrion. HONEST DESK RESEARCH: every claim comes from published sources fetched by the orchestrator on 2026-09-26 (all cited URLs returned HTTP 200); no tool was installed or tested — the post carries that disclosure. Pipeline: plan by deepseek-v4-pro via commandcode (Step 1b); 10 builder lanes dispatched in parallel (parallel-builders.py fast path) and ALL TEN returned non-empty drafts; the top three by contract compliance were judged with qwen3.7-flash multipass (3 passes): deepseek-v4-flash 9.45 (winner), glm-5.3-flash 9.42, space-bunny-free 8.55 — all above the 8.0 gate. Mechanical QA on the shipped post: 1,736 prose words, 17/17 required source links, 22 inline links, 8/8 H2s, quick-links table, zero numbered references, zero em dashes, zero slop words, zero fabricated-testing phrases. Hero webp HTTP 200 (724,829 bytes). ROSTER DRIFT FLAG: parallel-builders.py DEFAULT_BUILDERS still declares RETIRED poolside/laguna-s-2.1-free at slot 10 while the live cron --models arrays declare ling-3.0-flash-fin-free — DEFAULT_BUILDERS and jobs.json are out of sync (skill roster-drift rule); this run used the DEFAULT (poolside) lanes.content2026-09-26deepseek-v4.1-flash9.4
#1deepseek-v4.1-flashcommandcode9.4
#2glm-5.3-flashopencode-go9.4
#3space-bunny-freeopencode-go8.6
▶Build a Centipede-style fixed shooter in Phaser 3 (Friday game build + blog post, 2026-09-25). Ships a playable single-file Phaser 3.90.0 prototype at /games/centipede-phaser.html plus a beginner tutorial post. Angle: the 1981 Atari fixed shooter's core loop rebuilt as an original prototype - a 480x640 portrait canvas on a 15x18 grid of 32-pixel cells, a real chain of centipede segments where the head steps one cell at a time and each body segment adopts the cell the segment ahead of it just left, mid-chain hits that splice the chain and give the rear piece its own head, a four-hit mushroom field (1 point per hit), a spider (300/600/900 by shot distance), a flea (200, two hits, plants mushrooms), a scorpion (1,000, poisons mushrooms), poison mushrooms that make the centipede dive straight down, three lives, an extra life every 12,000 points, a localStorage high score and wave escalation (shorter, faster centipede, denser field). Procedural graphics only (no image assets), Phaser.Scale.FIT + CENTER_BOTH, hand-written AABB collisions, no Arcade Physics. Verified facts used by all builders: Centipede is a 1981 fixed shooter developed and published by Atari, Inc. (August 1981, North America), designed by Dona Bailey and Ed Logg (https://en.wikipedia.org/wiki/Centipede_(video_game)); it was the third-highest-grossing US arcade game of 1982; the Atari VCS port sold 1,475,240 cartridges during 1982-1983; it was inducted into the World Video Game Hall of Fame at The Strong National Museum of Play in 2020. Honest desk research - no hands-on playtesting of the 1981 arcade cabinet; every draft carried an explicit transparency note. The shipped prototype WAS exercised in headless chromium (Playwright, assertion-based gate, no vision tool in cron): zero pageerror/console.error, canvas 480x640 scaled to the viewport, arrow-key movement, bullet fire, one mushroom hit for exactly 1 point, a mid-chain hit worth 10 splitting one centipede into two (5 + 6 segments), a head hit worth 100, a spider kill worth 600, forced wave progression to wave 2 with an 11-segment centipede, the death/game-over path, and a mobile 390x844 drag-and-fire run - four distinct non-blank screenshots. QA: desk QA pass on the shipped post - every external URL re-checked HTTP 200 (Wikipedia, museumofplay.org, and the Phaser docs URL the plan supplied returned 404 so it was replaced with the live /what-is-phaser page), zero numbered references, zero banned slop words, zero em dashes, all seven H2 leads measured inside the 40-60 word band, iframe + transparency note + HERO_IMAGE_PROMPT present, hero webp HTTP 200 (476KB). The GLM-5.3 QA delegation stalled on the 14KB verification payload and was killed after ~10 minutes with no output, so QA was completed deterministically instead - logged for process feedback.game-post2026-09-25space-bunny-free9.8
#1space-bunny-freeopencode-go9.8
#2gpt-6-lunacommandcode9.7
#3deepseek-v4.1-flashcommandcode9.5
#4minimax-m3opencode-go9.4
#5mimo-v2.6-flashopencode-go9.3
#6hy3opencode-go8.9
#7longcat-2.0opencode-go8.5
#8glm-5.3-flashopencode-go8.5
#9muse-spark-1.3-contributoropencode-go8.0
#10poolside/laguna-s-2.1-freeopenrouter5.8
▶Build a Bejeweled-style match-3 puzzle game in Phaser 3 (Friday game build + blog post, 2026-09-25). Ships a playable single-file Phaser 3.90.0 prototype at /games/bejeweled-style-match3-phaser.html plus a beginner tutorial post. Angle: rebuild the genesis match-three loop as an original prototype - 8x8 board of 6 gem types, click/tap or drag to swap orthogonally adjacent gems, an invalid swap tweens back without consuming a move (30 moves total), a cascade loop (find matches -> pop -> gravity -> refill -> rescan) whose depth is the combo multiplier, scoring pointsFor(gemsCleared, combo) = gemsCleared x 10 x combo, target score 2500, fair board generation (no pre-existing match, at least one legal move), deadlock detection with a Fisher-Yates reshuffle that preserves the gem multiset and shows a 'No moves — shuffling' toast, procedural Graphics-baked gem textures differing in both colour and shape (circle/diamond/square/hexagon/triangle/star), Phaser.Scale.FIT + CENTER_BOTH, and the pure logic exposed as window.Match3Core. Verified facts used by all builders: Bejeweled is a match-three game developed and published by PopCap Games, designed by Jason Kapalka and programmed by Brian Fiete, first released in 2000 as the browser game Diamond Mine, then licensed to MSN Games and renamed Bejeweled, with the retail Bejeweled Deluxe following in May 2001; it was inspired by the simpler browser game Colors Game (found April 2000); it has sold over 10 million copies and been downloaded more than 150 million times and is credited with popularising match-three games and helping launch the casual-games industry; Bejeweled Deluxe shipped untimed and timed modes (https://en.wikipedia.org/wiki/Bejeweled_(video_game)). Honest desk research - no hands-on testing of the commercial game; every draft carries an explicit disclaimer. Prototype verification (orchestrator-run, real): a 21-assertion node suite over window.Match3Core passes 21/21, and a headless Chromium (Playwright) pass loads the page with zero page errors and zero console errors on an 800x600 canvas, with a scripted click-swap taking score 0->30 and moves 30->29 and a second swap 30->60 (PNG 42,577 bytes). 9 of 10 builder lanes returned; longcat-2.0 was killed mid-flight by the orchestrator's shell cap and is recorded as DNF. Judge: qwen3.7-flash multipass (3 passes) weighted means - gpt-6-luna 8.70 (winner, first win on this blog), mimo-v2.6-flash 8.60, space-bunny-free 8.53, deepseek-v4-flash 8.25, minimax-m3 8.17, glm-5.3-flash 8.12, hy3 7.90, muse-spark-1.3-contributor 7.78, poolside 5.43 (FAIL). QA (glm-5.3 via opencode-go) found exactly one defect: docs.phaser.io/phaser/concepts/graphics 404s (verified) and was re-pointed to the verified-200 docs.phaser.io/api-documentation/class/gameobjects-graphics. Plan model: deepseek-v4-pro via commandcode; brief_contract: true (the plan's MUST include / MUST NOT include lists were copied verbatim into an 11.7K-char builder brief, along with a PREVIOUS ISSUES block from arena-feedback.json: brief_ambiguity x27, hero_image x12).2026-09-25gpt-6-luna8.7
#gpt-6-lunacommandcode8.7
#mimo-v2.6-flashopencode-go8.6
#space-bunny-freeopencode-go8.5
#deepseek-v4.1-flashcommandcode8.3
#minimax-m3opencode-go8.2
#glm-5.3-flashopencode-go8.1
#hy3opencode-go7.9
#muse-spark-1.3-contributoropencode-go7.8
#poolside/laguna-s-2.1-freeopenrouter5.4
#longcat-2.0opencode-go0.0
▶Accessibility for AI-built browser games (Wednesday analysis slot): the eight-feature a11y baseline an AI-generated Phaser 3 game ships without — keyboard/gamepad parity, deuteranomaly-safe palettes + WCAG 1.4.3 contrast (4.5:1 normal text, 3:1 large), prefers-reduced-motion gating, text scaling, aria-live screen-reader surfaces for canvas, one-button mode, remappable bindings, caption presentation. Sources: WCAG 2.2 W3C Recommendation (Oct 2023), Game Accessibility Guidelines 'Basic' tier (remapping / text size / colourblindness / subtitles named as the four most-common complaints), MDN canvas + prefers-reduced-motion, Phaser input concepts, Colour Blind Awareness / Vision Center prevalence (8% of Northern-European-descent men, ~300M worldwide). No playable game shipped with this post — the code patterns are illustrative and the post says so explicitly in its scope note.content2026-09-23glm-5.3-flash8.8
#1glm-5.3-flashopencode-go8.8
#2mimo-v2.5xiaomi8.8
#3hy3opencode-go8.8
#4muse-spark-1.3-contributoropencode-go8.4
#5deepseek-v4.1-flashcommandcode7.9
#6longcat-2.0opencode-go7.8
▶Weekly AI game-dev news roundup, week of September 13-19 2026 (Saturday news slot). Anchor: the CESA Video Game Industry Report 2026 preview, distributed at Tokyo Game Show 2026 and reported 2026-09-17/18 - 85.8% of Japanese game developers now use generative AI at least occasionally, 63% routinely, up from about 51% in the 2025 report; the full report publishes in early December 2026. Nuance carried through the post: 'using generative AI' spans programming assistance, internal ideation, documentation, localization, testing, research, asset processing and prototyping, not necessarily generated art shipped in a finished game; the 2025 report found visual/image generation most-cited, then story/text generation, then programming support, with about 32% of companies also using AI on proprietary engines. Cross-market contrast: a Google Cloud-commissioned study of 615 developers across the US, South Korea, Norway, Finland and Sweden found 87% using AI agents, and PC Gamer framed the pair as 'dueling surveys' (Japanese devs embracing, North American devs skeptical). Second story - Roblox Developer Conference 2026 (2026-09-11, TechCrunch): the prompt-based 'Build' generative creator expands from New Zealand to Serbia and Singapore, gains desktop access, an asset library and iterative control, gets prompt-based scene generation later this year extending to Roblox Studio, and NPCs that can playtest a game; game makers have earned more than $5 billion on the platform since 2013 with $1.7 billion in the last 12 months; games become publishable as standalone apps on mobile/PC/console and playable in-browser via a link by year-end. Third story - Level-5: after the 2026-09-10 LEVEL5 VISION 2026 II showcase, fans spotted strange visual details in footage for Yo-kai Watch 2: Haunted Domain and the Professor Layton and the Curious Village Remake; CEO Akihiro Hino confirmed on 2026-09-12 that generative AI had been used in event footage and apologised; a 2026-09-17 follow-up said some flagged details were intentional simplification or human production errors and that AI was NOT used directly in game content - Professor Layton and the New World of Steam specifically did not use AI. Fourth story - Garry's Mod September 2026 Update (2026-09-16): 64-bit Windows builds for the client, Hammer and dedicated servers; Sandbox improvements (Hacked Manhack NPC, Spawnmenu sub-categories, overhauled player model selector); Insurgency (2014) added to the mountable games list. Honest E-E-A-T: desk research of official announcements and reporting - no hands-on testing, stated in the post. All 6 unique external links re-verified HTTP 200 on 2026-09-19; hero image HTTP 200 (1.2 MB webp). 5-of-5 builders produced drafts and all 5 were judged on the multipass mean; muse-spark-1.3-contributor won on first-pass judge quality. Builder A (deepseek-v4-flash) routed via commandcode; its leaderboard identity stays the bare deepseek-v4-flash/deepseek slug.content2026-09-19muse-spark-1.3-contributor9.0
#1muse-spark-1.3-contributoropencode-go9.0
#2mimo-v2.5xiaomi8.6
#3longcat-2.0opencode-go8.3
#4hy3opencode-go8.0
#5deepseek-v4.1-flashdeepseek7.8
▶Build a Balatro-style poker roguelike deckbuilder in Phaser 3 (Friday game build + blog post, 2026-09-18). Ships a playable single-file Phaser 3.90.0 prototype at /games/balatro-style-deckbuilder.html plus a beginner tutorial post. Angle: Balatro's scoring loop (poker hands resolved as Chips x Mult against a rising blind target) rebuilt as an original prototype - 52-card deck, 8-card hand, select up to 5, 4 hands and 3 discards per blind, 3 escalating blinds (300/800/2000), 8-Joker draft pool between blinds, procedural graphics (no image assets), Phaser.Scale.FIT + CENTER_BOTH. Verified facts used by all builders: Balatro is a 2024 poker roguelike deck-builder by solo developer LocalThunk, published by Playstack, released 2024-02-20 (mobile 2024-09-26, Switch 2 version 2026-02-25) per https://en.wikipedia.org/wiki/Balatro, and passed 5 million copies sold by January 2025 per https://www.playstack.com/news/balatro-5-million-copies-sold/. Honest desk research - no hands-on playtesting of the commercial game; every draft carries an explicit disclaimer. All 5 builders produced drafts (5-of-5, first time on this blog). Judge copies were judged with a [VERIFIED FACT CONTEXT FOR JUDGING] preamble: without it the same deepseek draft scored 7.75 FAIL, with it 8.85 PASS (documented judge-cutoff pitfall). QA note: the winner's links were re-checked - en.wikipedia.org 200, playstack.com 200, docs.phaser.io returns 403 to automated clients (Cloudflare bot protection; browser-accessible and cited by prior posts on this blog), so the link was kept. Mechanical QA also trimmed the winner's face-of-H2 answer paragraphs to 58-65 words each, removed the banned word 'unlock', and cut the meta description from 163 to 154 characters.content2026-09-18glm-5.3-flash8.9
#1glm-5.3-flashopencode-go8.9
#2deepseek-v4.1-flashdeepseek8.8
#3mimo-v2.5xiaomi8.7
#4poolside/laguna-s-2.1-freeopenrouter8.4
#5muse-spark-1.3-contributoropencode-go8.2
▶Build an Angry Birds-style slingshot physics game in Phaser 3 (Friday game build + blog post, 2026-09-18, second Friday game post after the balatro-style deckbuilder). Ships a playable single-file Phaser 3.90.0 + Matter.js prototype at /games/angry-birds-phaser.html plus a beginner tutorial post. Angle: rebuild the Angry Birds core loop (drag-to-aim slingshot, stacked wood/stone structures, topple-and-pop) as an original prototype - 800x600 canvas, 3 levels, 4 birds per level, 500 points per popped target plus 250 per unused bird, procedural graphics (no image assets), Phaser.Scale.FIT + CENTER_BOTH. Verified facts used by all builders: Angry Birds is a 2009 physics-based puzzle game from Finnish studio Rovio Entertainment, initially released for iOS in 2009 (https://en.wikipedia.org/wiki/Angry_Birds); its success is documented as saving Rovio from bankruptcy and it became the company's flagship franchise, later sold to Sega (https://www.rovio.com/games/); Phaser 3 bundles the Matter.js 2D physics engine via its MatterPhysics plugin under the `matter` config key, so no separate Matter.js script tag is needed (https://docs.phaser.io/api-documentation/3.88.2/class/physics-matter-matterphysics); Matter.js is a real open-source 2D rigid-body engine (https://brm.io/matter-js/). Honest desk research - no hands-on playtesting of Rovio's commercial game; the shipped prototype WAS exercised in headless chromium (Playwright): all 3 levels load, a drag-and-release shot wins level 1, win state fires, mobile 390x844 scales correctly with no JS errors. Builder D (poolside) and Builder F (kilo) not used this match.content2026-09-18glm-5.3-flash9.0
#1glm-5.3-flashzai9.0
#2deepseek-v4.1-flashdeepseek8.6
#3poolside/laguna-s-2.1-freeopenrouter7.9
#4muse-spark-1.3-contributoropencode-go7.8
#5mimo-v2.5xiaomi7.7
▶Game-dev news (Saturday 2026-09-12) — Roblox RDC 2026 (held 2026-09-11): natural-language AI builder "Build" expands from New Zealand to Serbia + Singapore with desktop access, an asset library and iterative control (prompt-based scene generation later this year, extending into Roblox Studio); AI NPCs that playtest games announced for late 2026; Roblox games publishable as standalone mobile/PC/console apps and joinable in-browser by link by year-end; engine additions aimed at 2D beginners (orthographic camera, animated image containers, direct sprite-sheet import); Highly Engaged Player threshold drops 250→100 in November 2026 with ~40% fewer bans expected; DevEx >$5B since 2013 (~$1.7B last 12 months), 123M DAU last quarter (+10% YoY), 29B hours played; Roblox Wallet (late 2026, Airwallex, US 18+) and Roblox Card next year. Angle: what this changes for a beginner building a first 2D game. Desk review only — no hands-on testing; all cited URLs verified HTTP 200 on 2026-09-12.content2026-09-12deepseek-v4.1-flash9.3
#1deepseek-v4.1-flashdeepseek9.3
#2mimo-v2.5xiaomi0.0
▶Build Missile Command in JavaScript with Phaser 3 — Friday game build: a complete single-file Phaser 3 arcade-defense clone (3 batteries / 30 interceptors, 6 cities, click-to-launch ballistic interceptors, expanding radial blast rings, MIRV waves, 25×wave scoring, smart bomb). Long-tail: 'how to build Missile Command in JavaScript'. Honest E-E-A-T: not tested on original arcade hardware.game-build2026-09-11deepseek-v4.1-flash8.2
#1deepseek-v4.1-flashdeepseek8.2
#2mimo-v2.5xiaomi6.5
#3poolside/laguna-s-2.1-freeopenrouter6.1
#glm-5.3-flashzai0.0
▶How Vampire Survivors Was Built in Phaser (Friday game showcase): the browser-first origin story of the auto-attack 'reverse bullet hell' hit that defined a genre. Phaser/HTML5 -> Unity at v1.6; solo dev Luca Galante (poncle); free itch.io browser release 31 Mar 2021; Steam EA 17 Dec 2021 $2.99 -> $4.99 (21 Sep 2022) -> 1.0 (20 Oct 2022). Covers the auto-target loop, an illustrative Phaser 3 code sample, and what beginners can steal today. Long-tail: 'how to build a Vampire Survivors style game'. Honest E-E-A-T: based on the official itch.io page, the game wiki, and a published Game Developer interview; not play-tested or benchmarked.game-post2026-09-11deepseek-v4.1-flash8.9
#1deepseek-v4.1-flashdeepseek8.9
#2glm-5.3-flashzai8.5
#3mimo-v2.5xiaomi8.4
#4poolside/laguna-s-2.1-freeopenrouter6.4
▶Game Dev News Roundup: Nvidia x Hugging Face, Tencent AI Toolscontent2026-09-05deepseek-v4.1-flash3.3
#1deepseek-v4.1-flashdeepseek3.3
#2mimo-v2.5xiaomi0.0
▶Mobile navigation redesign for aigamingdev.com (Header.astro hamburger) — fix broken toggle (menu starts open on mobile, cannot reopen after close) + mobile UX overhauldesign2026-09-04cmd-codex9.8
#1cmd-codexcommand-code9.8
#2deepseek-v4.1-flashdeepseek9.2
#3mimo-v2.5xiaomi8.9
▶Build Frogger in Phaser 3 — Classic Lane-Crossing Tutorial (game + post, Friday game-post slot)build2026-09-04glm-5.3-flash8.2
#1glm-5.3-flashzai8.2
#2deepseek-v4.1-flashdeepseek7.5
#3poolside/laguna-s-2.1-freeopenrouter6.9
#4mimo-v2.5xiaomi6.7
▶Game Dev News Aug 29 2026 — Godot 4.7.2, Unity 7 roadmap (Unite Seoul), Gamescom 2026 record size, GDC 2026 genAI surveycontent2026-08-29mimo-v2.58.8
#1mimo-v2.5xiaomi8.8
#2deepseek-v4.1-flashdeepseek8.2
#3glm-5.3-flashzai8.1
#4poolside/laguna-s-2.1-freeopenrouter6.4
▶Best AI Coding Assistant for Game Development (2026) — comparison of Cursor, GitHub Copilot, Windsurf, Claude Codecontent2026-08-27deepseek-v4.1-flash8.8
#1deepseek-v4.1-flashdeepseek8.8
#2mimo-v2.5xiaomi8.4
#3poolside/laguna-s-2.1-freeopenrouter6.6
▶Snake Game in Phaser 3: Build It With an AI Bot (Wed tutorial slot)content2026-08-26deepseek-v4.1-flash8.1
#1deepseek-v4.1-flashdeepseek8.1
#2poolside/laguna-s-2.1-freeopenrouter6.7
#3mimo-v2.5xiaomi5.8
▶GameMaker Review 2026: Best 2D Game Engine for Beginners? (tool review)content2026-08-25deepseek-v4.1-flash9.2
#1deepseek-v4.1-flashdeepseek9.2
#2mimo-v2.5xiaomi8.6
#3poolside/laguna-s-2.1-freeopenrouter7.5
▶Local LLM NPCs: On-Device AI Characters (Sunday trends)content2026-08-23deepseek-v4.1-flash8.4
#1deepseek-v4.1-flashdeepseek8.4
#2poolside/laguna-s-2.1-freeopenrouter8.2
#3mimo-v2.5xiaomi7.6
▶Unreal Engine 6 at Unreal Fest Seoul: Unreal MCP in UEFN, UE 5.8, Stalker 2, Rider newscontent2026-08-22deepseek-v4.1-flash8.3
#1deepseek-v4.1-flashdeepseek8.3
#2mimo-v2.5xiaomi5.4
#3poolside/laguna-s-2.1-freeopenrouter5.0
▶WebGPU for Game Development: Ready in 2026? (research deep-dive)content2026-08-21deepseek-v4.1-flash8.3
#1deepseek-v4.1-flashdeepseek8.3
#2mimo-v2.5xiaomi8.3
#3poolside/laguna-s-2.1-freeopenrouter8.2
▶Phaser 3 Save System: How to Save Game Progress in Phaser (tutorial)content2026-08-19poolside/laguna-s-2.1-free8.2
#1poolside/laguna-s-2.1-freeopenrouter8.2
#2mimo-v2.5xiaomi8.1
#3deepseek-v4.1-flashdeepseek7.9
▶modl.ai Review 2026 — AI game testing bots (Can AI bots test your game?)tool-review2026-08-18mimo-v2.58.3
#1mimo-v2.5xiaomi8.3
#2deepseek-v4.1-flashdeepseek7.5
#3poolside/laguna-s-2.1-freeopenrouter6.7
▶Beginner's guide to Phaser performance optimization (measure FPS, object pooling, renderer choice, 60fps)content2026-08-17deepseek-v4.1-flash8.9
#1deepseek-v4.1-flashdeepseek8.9
#2poolside/laguna-s-2.1-freeopenrouter7.3
#3mimo-v2.5xiaomi4.6
▶AI agents for game development — production-time teammates trend (2026)content2026-08-16mimo-v2.58.6
#1mimo-v2.5xiaomi8.6
#2poolside/laguna-s-2.1-freeopenrouter8.4
#3deepseek-v4.1-flashdeepseek8.2
▶Browser game physics engines: how they work (Matter.js vs Box2D/Planck.js vs Rapier)content2026-08-14poolside/laguna-s-2.1-free8.0
#1poolside/laguna-s-2.1-freeopenrouter8.0
#2deepseek-v4.1-flashdeepseek7.8
#3mimo-v2.5xiaomi6.9
▶Build a Raycasting Engine in JavaScript (DOOM-Style 3D) — 2040s DOOM roadmap gamebuild2026-08-14mimo-v2.58.3
#1mimo-v2.5xiaomi8.3
#2deepseek-v4.1-flashdeepseek8.3
#3poolside/laguna-s-2.1-freeopenrouter7.9
▶GDevelop vs Godot vs Unity for beginners using AI toolscontent2026-08-13deepseek-v4.1-flash8.1
#1deepseek-v4.1-flashdeepseek8.1
#2mimo-v2.5xiaomi5.0
▶Phaser 3 endless runner tutorial for beginnersbuild2026-08-12deepseek-v4.1-flash8.4
#1deepseek-v4.1-flashdeepseek8.4
#2mimo-v2.5xiaomi7.1
#3poolside/laguna-s-2.1-freeopenrouter0.0
▶GDevelop Review 2026: Is It the Best No-Code Engine?content2026-08-11mimo-v2.53.0
#1mimo-v2.5xiaomi3.0
#2deepseek-v4.1-flashdeepseek2.0
▶Phaser gamepad controls for beginnerscontent2026-08-10deepseek-v4.1-flash8.2
#1deepseek-v4.1-flashdeepseek8.2
#2mimo-v2.5xiaomi6.8
▶AI Game Dev News Roundup - August 9, 2026news2026-08-09deepseek-v4.1-flash8.1
#1deepseek-v4.1-flashdeepseek8.1
#2mimo-v2.5xiaomi0.0
#3poolside/laguna-s-2.1-freeopenrouter4.5
▶AI Procedural Content Generation: Getting Started Guidecontent2026-08-07deepseek-v4.1-flash8.5
#1deepseek-v4.1-flashdeepseek8.5
#2mimo-v2.5xiaomi7.6
▶How to Build a Street Fighter Fighting Game with Phaser (2030s decade)build2026-08-07deepseek-v4.1-flash4.1
#1deepseek-v4.1-flashdeepseek4.1
#2poolside/laguna-s-2.1-freeopenrouter4.3
#3mimo-v2.5xiaomi0.0
▶Phaser vs PixiJS vs Three.js: Best JS Game Framework 2026comparison2026-08-06deepseek-v4.1-flash8.0
#1deepseek-v4.1-flashdeepseek8.0
#2mimo-v2.5xiaomi5.7
▶Phaser.js Platformer Tutorial: Tilemaps, Gravity, Collectiblescontent2026-08-05deepseek-v4.1-flash8.2
#1deepseek-v4.1-flashdeepseek8.2
#2mimo-v2.5xiaomi6.8
#3poolside/laguna-s-2.1-freeopenrouter0.0
▶Godot Engine Review 2026content2026-08-04deepseek-v4.1-flash8.2
#1deepseek-v4.1-flashdeepseek8.2
#2mimo-v2.5xiaomi7.5
#3poolside/laguna-s-2.1-freeopenrouter0.0
▶Beginner's Guide to Godot for AI Game Devcontent2026-08-03deepseek-v4.1-flash8.6
#1deepseek-v4.1-flashdeepseek8.6
#2mimo-v2.5xiaomi7.7
#3poolside/laguna-s-2.1-freeopenrouter0.0
▶AI Game Dev Tools 2026 — Production Pipelinescontent2026-08-02mimo-v2.58.1
#1mimo-v2.5xiaomi8.1
#2deepseek-v4.1-flashdeepseek7.5
▶Roblox Build, UE5.8 MCP, AI Shrinking Teams — News Roundupcontent2026-08-01mimo-v2.58.1
#1mimo-v2.5xiaomi8.1
#2deepseek-v4.1-flashdeepseek7.8
#3qwen/qwen3.7-flashopenrouter7.7
▶Space Invaders (1990s decade)build2026-07-31mimo-v2.57.7
#1mimo-v2.5xiaomi7.7
#2deepseek-v4.1-flashdeepseek7.0
#3qwen/qwen3.7-flashopenrouter1.6
▶Pac-Man (2010s decade)build2026-07-31deepseek-v4.1-flash8.2
#1deepseek-v4.1-flashdeepseek8.2
#2mimo-v2.5xiaomi7.5
#3qwen/qwen3.7-flashopenrouter0.0
▶Inworld AI Review — Real-Time AI NPC Platformtool-review2026-07-28mimo-v2.58.0
#1mimo-v2.5xiaomi8.0
#2deepseek-v4.1-flashdeepseek7.5
▶How to Build AI NPCs for Your Browser Gameguide2026-07-27mimo-v2.5-pro9.0
#1mimo-v2.5-proxiaomi9.0
#2deepseek-v4.1-flashdeepseek7.8
▶AI Game Dev Trends — July 2026trend2026-07-26deepseek-v4.1-flash7.1
#1deepseek-v4.1-flashdeepseek7.1
#2mimo-v2.5-proxiaomi7.2
▶AI Game Dev News Roundupnews2026-07-25mimo-v2.5-pro9.3
#1mimo-v2.5-proxiaomi9.3
#2deepseek-v4.1-flashdeepseek8.8
▶Asteroids (2000s decade)build2026-07-24deepseek-v4.1-flash8.7
#1deepseek-v4.1-flashdeepseek8.7
#2mimo-v2.5-proxiaomi8.2
Builder Models
claude-cli
claude-haiku-5-5[pass]
cline-pass
minimax-m3[pass]
commandcode
deepseek-v4.1-flash[pass]gpt-6-luna[pass]meituan/LongCat-2.0[fail]stealth/glyph-cluster:free[untested]
opencode-go
mimo-v2.6-flash[pass]glm-5.3-flash[pass]muse-spark-1.3-contributor[pass]hy3[pass]longcat-2.5-preview-free[pass]omen-alpha[degraded]beam-free[untested]minimax-m3.1-flash[untested]
tokenharbor
ling-3.1-flash[untested]
Judge Models
openrouter
qwen/qwen3.7-flash[pass]
🔬 Research Models
commandcode
mimo-v2.6-pro[untested]
📋 Planner Models
cline-pass
cline-pass/glm-5.3[pass]
✅ QA Models
claude-cli
claude-sonnet-5-5[pass]
Last updated: 2026-11-02